July 09, 2017

img/tree.jpg

img/title.png

Context

  • Mozambique is among the 10 countries with the greatest malaria burden.
  • Prevalence as high as 40% in some regions.
  • 45% of all outpatient cases.
  • 56% of all inpatient cases.
  • 26% of all hospital deaths.

  • 100 deaths daily.

  • Bordering countries in elimination phase.

img/border.png

Mozambican Alliance Towards the Elimination of Malaria

img/maltem.png

Timeline

img/time_line.png

Objectives

  1. Impact on malaria incidence (burden)

  2. Impact on labour force (absenteeism and productivity)

3. Impact on school age children (absenteeism and performance)

img/objectives.png

Literature

  • Early childhood health shocks -> long-term economic outcomes (Smith 2009, Case et al 2002 & 2005, Currie and Stabile 2003)
  • Malaria shown to have lifelong effects on economic outcomes in children - poor cognitive ability, higher absenteeism, poor labour market outcomes (Sachs and Malaney 2002, Lucas 2010)
  • Malaria infections among adults (Cutler 2010) shown to lower household income, worsen nutrition and schooling, restrict labour opportunities

Why is our study interesting?

We're different

  • Previous studies are ecological, retrospective, long-term
  • Ours is person-level, real-time, short-term

Potential impact

  • Elimination campaigns: very easy to estimate cost; very hard to estimate benefit
  • Short-term trumps the long-term for most policy-makers

Identification strategy

  • We compare grades of students from selected schools in Intervention (Magude district) and Control (Manhiça district) areas
  • For both intervention and control we collected data for the year before the intervention started (year 2015) and for the year after the intervention started (year 2016).
  • We apply difference in difference analysis to identify and measure the short term impact of the elimination campaign

Study area

Similarity across districts

Characteristic Manhiça Magude
Age mean 9.38 9.20
Age SD 2.70 2.81
Education mean 2.40 2.51
Education SD 1.19 1.40
Females 2777.00 899.00
Males 2869.00 877.00
N 7671.00 3693.00
Ses asset score mean 6972.72 6770.72
Ses asset score SD 961.59 852.05
Siblings mean 4.48 4.41
Siblings SD 2.52 2.85

Econometric models

Difference in differences (1)

In diff-diff models: assumption: parallel trends (in the outcome variable) in treatment and control groups before the introduction of the intervention In order to check this crucial assumption…

img/formula1.png

Event study (2)

The coefficients for the interaction between trimester and intervention area identify the differential trends in the outcome variable over time between the treated and control regions. Thus, as the policy is implemented in 2016, these coefficients identify any differential pre-trends in the outcome variable between treated and control regions (the crucial assumption for diff-diff models).

img/formula2.png

Data collection

  • 4 Schools in Magude (Intervention); 5 Schools in Manhiça (Control)
  • Pre-intervention: year 2015
  • Post-during intervention: year 2016

img/pauta.png

Data collection

img/pauta2.png

Descriptive statistics

  • 9 schools (5 intervention, 4 control)
  • 140 classes (across the 2 years observation)
  • 8,832 students
  • 222,503 observations (on trimester examination)
  • 860,927 observations (on daily absenteeism) (unbalanced panel)

Absenteeism

Average absenteeism rate, per term, per school

Flat-line vs. decline

Grades

Density curves by subject

Density cuvres by subject and district

Density cuvres by subject, year, and district

Distribution of grades, all subjects

Just maths

Proportion of those passing exams (grade >= 10/20), all subjects

Proportion passing exams table (>= 10), trimester level

Proportion passing exams table (>= 10), trimester level, math only

Mean grade obtained, all subjects

Mean grade obtained, all subjects, by trimester

Mean grade obtained, math only, by trimester

Regressions

Passed trimester exam (all subjects)

Regression versions
Term Key 1 2 3
After Estimate 0.002 0.003 0.003
P 0.128 0.019 0.019
S.E. 0.002 0.001 0.001
After:Intervention Estimate 0.021 0.02 0.02
P < 0.001 < 0.001 < 0.001
S.E. 0.003 0.003 0.003
(Intercept) Estimate 0.867 0.963 0.964
P < 0.001 < 0.001 < 0.001
S.E. 0.002 0.002 0.002
Intervention Estimate 0.044 0.036 0.036
P < 0.001 < 0.001 < 0.001
S.E. 0.005 0.005 0.005

All regressions controlling for school. Regressions II and III also controlling for subject.

**The impact of the policy is to increase the probability of passing the exam by 2 percentage points. Given that the proportion of those passing examinations was 87.5% in the intervention area in 2015, the increase due to the intervention is (0.02/0.875*100) 2.28%.**

Passed trimester exam (Maths only)

Term Key value
After Estimate 0.006
S.E. 0.005
P 0.253
factor(trimester)2 Estimate < 0.001
S.E. 0.005
P 0.003
factor(trimester)3 Estimate < 0.001
S.E. 0.005
P < 0.001
After:Intervention Estimate 0.052
S.E. 0.01
P < 0.001
(Intercept) Estimate 0.786
S.E. 0.007
P < 0.001
Intervention Estimate 0.026
S.E. 0.018
P 0.143

The impact of the policy is to increase the probability of passing the exam by 5 percentage points for the case of maths. Given that the proportion of those passing examinations was 76.69% in the intervention area in 2015, the increase due to the intervention is (0.05/0.7669*100) 6.52%.

OLS regression; regression controlling for school (coeff not shown);

Average grade, all subjects

Regression versions
Term Key 1 2 3
After Estimate 0.169 0.175 0.175
P < 0.001 < 0.001 < 0.001
S.E. 0.014 0.014 0.014
After:Intervention Estimate 0.233 0.229 0.229
P < 0.001 < 0.001 < 0.001
S.E. 0.026 0.026 0.026
(Intercept) Estimate 11.822 12.18 12.167
P < 0.001 < 0.001 < 0.001
S.E. 0.015 0.02 0.022
Intervention Estimate 0.017 < 0.001 < 0.001
P 0.72 0.507 0.509
S.E. 0.048 0.047 0.047

The impact of the policy is to increase the grade for all subjects by 0.23 percentage points. Given that the mean grade was 12.11 in the intervention area in 2015, the increase due to the intervention is (0.22/12.11*100) 1.9%.

Average grade (Maths only)

Term Key value
After Estimate 0.153
S.E. 0.046
P < 0.001
factor(trimester)2 Estimate < 0.001
S.E. 0.047
P < 0.001
factor(trimester)3 Estimate < 0.001
S.E. 0.048
P < 0.001
After:Intervention Estimate 0.529
S.E. 0.088
P < 0.001
(Intercept) Estimate 12.087
S.E. 0.058
P < 0.001
Intervention Estimate < 0.001
S.E. 0.158
P 0.005

Event study

Event study model

Term Key value
(Intercept) Estimate 0.965
Estimate 0.056
-1 Period Estimate < 0.001
1 Period Estimate 0.004
-2 Period Estimate < 0.001
2 Period Estimate < 0.001
-3 Period Estimate 0.002
-1 Time::Intervention Estimate < 0.001
1 Time::Intervention Estimate 0.002
-2 Time::Intervention Estimate < 0.001
2 Time::Intervention Estimate < 0.001
-3 Time::Intervention Estimate < 0.001

Event study model - Maths only

Term Key value
(Intercept) Estimate 0.783
S.E. 0.007
P < 0.001
Estimate 0.086
S.E. 0.02
P < 0.001
-1 Period Estimate < 0.001
S.E. 0.009
P 0.003
1 Period Estimate < 0.001
S.E. 0.009
P 0.82
-2 Period Estimate < 0.001
S.E. 0.009
P 0.02
2 Period Estimate < 0.001
S.E. 0.009
P 0.159
-3 Period Estimate 0.015
S.E. 0.009
P 0.092
-1 Time::Intervention Estimate < 0.001
S.E. 0.017
P 0.004
1 Time::Intervention Estimate < 0.001
S.E. 0.016
P 0.874
-2 Time::Intervention Estimate < 0.001
S.E. 0.017
P 0.001
2 Time::Intervention Estimate < 0.001
S.E. 0.016
P 0.207
-3 Time::Intervention Estimate < 0.001
S.E. 0.017
P < 0.001

Event study visualization

Discussion / Conclusion

  • Preliminary results point to a positive short term impact of the malaria campaign on school performance, both in terms of grades as continutous variable and passing examinations;
  • The channel through which the campaign had an impact on school performance is absenteeism reduction, the next outcome to investigate;
  • Next steps:
  • controlling for students’ characteristics: sex, socio-economic conditions, etc
  • Investigating issues related to the multilevel nature of database
  • Investigating issues related to unbalanced panels

Obrigado!

img/tree.jpg

Additional slides

Contamination effect map

Contamination map 2

Showing the difference between 2015 and 2016 average grades

Contamination chart - school performance

Contamination chart - absenteeism

Contamination - student-level

Descriptive statistics on performance

  • 9 schools
  • 8832 students
  • 222503 observations (subject-specific trimester exams)
  • Unbalanced panel

Number of unique students, per year in performance

Number of unique students, per year, per district

Number of students followed through both years

2162 of 5162 (41.88%)

Average number of observations, per student, per year

(One observation = 1 trimester-class)

Total number of observations, per year

(One observation = 1 trimester-class)

Average yearly grade, per school and year

Absenteeism

Number of unique students, per year

Number of unique students, per year, per district

Number of students followed through both years

1737 of 4662 (37.26%)

Average number of observations, per student, per year

(One observation = 1 student-day)

Total number of observations, per year

(One observation = 1 student-day)

Combined

Number of students that we have both info on absenteeism and performance

Total magnitude of database

  • 860927 student-days observed (absenteeism)

  • 222503 student-class-trimesters observed (performance)

Average absenteeism rate, per term, per district

Additional charts

Sex

Sex and absetneeism over time

Distance to school

Distance to school

Distance to school

Number of siblings

Number of siblings

Education

Education and absenteeism

Education and absenteeism

Education and absenteeism over time

SES Assets

Assets

Assets

Assets

Conditions

Performance over time

Performance over time

Performance over time

Performance over time

Performance over time

Performance over time

Performance over time

Other indictors

Monthly absenteeism by district

Absenteeism by term and district

Absenteeism by year and district

Standardized absenteeism